A Mysterious GPU Startup Sets Up In The Middle East

0:00Today on the AI Breakdown, we're looking at Stack Overflow, Intel, Spotify, and DoorDash, all adding AI features. Before that on the brief, a mysterious GPU startup is getting set up in the Middle East. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our Discord, our newsletter, and our YouTube channel. Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. Today we are starting with one of the absolute biggest themes of the year, and that is, of course, the AI chip shortage.

0:37This has shown up everywhere. It's impacted what companies like OpenAI have been able to release. They went into this year thinking they were going to be able to roll out multimodal features, for example, but that, among other things, has been cut off by lack of access to compute. Now, of course, on the flip side, the lack of access to compute has meant that the companies who can actually deliver it have done extraordinarily well. NVIDIA, for example, is up 160 % on the year as just the most easy example of this. But because there is such a clear shortage here, everyone is always interested when there are inklings of meaningful efforts to attempt to come at that in different ways.

1:15Over the last 24 hours or so, there has been a fair bit of chatter around this reporting from the information discussing what they call a mysterious AI data center startup hiring from AWS, Azure, and Meta. The information sets up the dynamics of the industry right now, saying, the cloud incumbents, including Microsoft and Google, are selling or developing their own AI server chips to compete with NVIDIA's graphics processing units, or GPUs. So NVIDIA is eager to prop up competition and flex its new king-making abilities by giving its latest GPU, the H100, to cloud server startups such as CoreWeave, Lambda Labs, and Crusoe Energy.

1:50They go on. Seemingly every day we hear about new companies that want to join the list to get into the GPU server rental bonanza. But then this is where it gets really interesting. We're also hearing a lot about a stealth GPU server rental startup that plans to operate among other places in the Middle East, where power is plentiful and can be relatively inexpensive. Sources tell us several well-known cloud computing and data center veterans have joined the startup. Sean Boyle, who served as chief financial officer of AWS until 2020. Gushag Ravade, a former Microsoft vice president and distinguished engineer who worked on its Azure server business until 2021, and T.S.

2:24Khurana, Meta Platform's vice president of infrastructure until last month. One person who was recruited by the startup said the company told them it has backing from a Kuwaiti royal family member and hopes such a connection will help it secure cheap power to run its data center servers. This person wasn't told the company's name, but according to one of Boyle's online profiles, the name appears to be Omniva. Now from there, the information tries to suss out what it can. They say that the company might be trying to develop their own server cooling technique to handle those advanced chips, basing that off of job postings they've seen.

2:53They'd also discuss why the Middle East is such a hotbed for this type of activity, and how the UAE has tried to become a significant player in the race to develop these new LLMs. And where they land is this. The race among startups to win NVIDIA's Heart and H100 chips is only beginning, and expect to see a lot more action involving GPUs in the Middle East. Next up today, the antagonists from other areas of big tech in the government are bringing their antagonism to AI as well. SEC Chair Gary Gensler is getting more and more vocal about his opinions on AI, taking a break from his crypto-killing schedule, with the latest that the SEC is proposing new rules around how Wall Street firms can use AI or not to attract clients.

3:32Basically, there's a line of thinking that suggests that the meme stock movement was all about online brokers and robo-advisors using, quote, game-like features to drive user behavior. A lot of reasonable people can disagree on whether or not that's true, but Gensler and his SEC are making their feelings clear. Now, if Gensler has been the main opponent of the crypto set, when it comes to big tech in general for this White House, that role has been played by FTC chair Lena Kahn. Kahn told CBS News, quote, We are seeing risks that AI could be used to turbocharge fraud and scams. We are also looking to be vigilant that we don't see anti-competitive practices or unfair methods of competition where some of the larger firms that have an advantage in this market are not using that power to squash competition.

4:12Now, ultimately, these issues, both those brought up by Khan and by Gensler, are important when it comes to AI. The question is just whether these particular voices around these issues are a little too political to actually make good progress on them. Reuters is reporting that revenue from Meta's Reels feature is, as they put it, narrowing in on TikTok boosted by AI. The piece begins, Reels, Meta Platform's answer to viral short-form video app TikTok, elicited eye rolls when it launched in 2020 and was regarded as yet another example of Meta copying a popular rival. But on Wednesday, Meta revealed numbers that show Reels videos are growing rapidly among both users and advertisers and are quickly catching up to the ByteDance-owned TikTok app that is beloved by young users and has reshaped the social media landscape.

4:52The article says that the number of Reels that are played on Facebook and Instagram has jumped from$140 billion last fall to$200 billion per day now. Alongside that, the annual revenue run rate has jumped to$10 billion, up from$3 billion last fall. That puts it right around the size of TikTok's business last year, and just a little behind and forecasts of TikTok's revenue this year, which analysts estimate to be a little above$13 billion. As Reuters puts it,

5:21Lastly,

5:27a fun little one from the art world. Some art experts are claiming that AI has proven conclusively that a controversial painting is in fact the work of old master Raffaul. Raphael. The Debrese Tondo is a piece whose provenance has been in question for the last 40 years since it was purchased at auction. Earlier this year, researchers at the University of Nottingham and Bradford said that an AI-powered facial recognition model suggested that there was a 97 % similarity between the Virgin Mary that was depicted in this disputed painting and a painting that was a confirmed Raphael called the Sistine Madonna.

6:00That AI tool also said there's an 86 % similarity between the child and the altarpiece's Jesus. Hassan Ugell, the professor of visual Computing at the University of Bradford who developed the model, said basically that the art world has been entirely unscientific. He said, it has been quite a learning curve to understand the art world and how little they use scientific evidence. The model, which he developed, uses, quote, dimensions the human eye can't see. Rudolf Hiller von Gertrigen, an art historian at the University of Leipzig who wrote a book about Raphael, said that he doesn't think that the painter would have copied himself effectively, that at the height of his career he preferred to various subjects, meaning it's much more likely a later reproduction or imitation.

6:37An Italian Renaissance expert at the University of New Hampshire, Patricia Emerson, said much the same. To repaint the Madonna and Child motif that one sees in the Sistine Madonna altarpiece is beneath his artistic dignity, he's not just looking for the extra buck. Beyond just being skeptical of this diagnosis, they are skeptical of AI models in general. One expert said, If you're deeply immersed in an artist's work, you've read everything about them. You've been to all the museums all over the world to see originals. You've been to gallery exhibitions. Maybe you've owned a few or bought and sold them.

7:04I don't think that sort of thing can be taught. But whether it can be taught or not, the stakes are certainly high. The experts quoted in the piece estimate that a new old master painting being rediscovered could fetch nine figures at auction. And so the$100 million question remains. Is this a real Raphael? I guess we're going to need GPT-5 for that. If you're enjoying the AI Breakdown Brief, please subscribe to it wherever you best consume content. And I'll be back soon with the main AI breakdown. Before we get into the main AI breakdown, I want to tell you about today's sponsor, Supermanage.

7:37If you work in a professional setting, you probably have some version of a one-on-one meeting, either with the people that work for you or the people that you work with. Unfortunately, all too often, those one-on-one meetings become glorified catch-up calls. Don't you wish you could jump right to the stuff that really matters? That's where Supermanage comes in. Supermanage AI magically distills your team's public Slack channels into a real-time brief on any employee, any time. Catch up on contributions, work in progress, challenges they're facing, sentiment, everything you need to show up ready for a truly meaningful conversation.

8:10And it's completely free. Visit supermanage.ai forward slash breakdown today to start making the most of your one-on-ones. And thanks again to Supermanage for sponsoring the AI breakdown. Today, my friends, we are answering one of the most burning questions in the modern economy. is AI hot? Welcome back to the AI Breakdown. Today we are effectively doing a case study of just how much every company in the world across basically all categories is having to figure out how they are going to incorporate AI into their offerings. Now this does tend to happen anytime there's some big tech movement, right?

8:50There were, for example, many companies last cycle talking about what their metaverse strategy would be. However, for one who watches these sort of frontier tech movements and hype cycles, it is very clearly different right now. And while it is highly unlikely that some of these nascent efforts, for example, we'll talk about DoorDash's order assistant chatbot in just a moment, will ultimately be the thing that really works for a particular industry, it seems highly likely that artificial intelligence powered tools, be they LLMs or some other type of generative AI are going to find their way into just about every type of business.

9:25So today we're going to go over four or five case studies of companies putting AI into their products, all from the last 24 hours of news. We begin with Stack Overflow. Yesterday, countless developers excitedly tweeted things like Danny Thompson here who says, Stack Overflow just announced their own AI. And unlike many of the companies that we'll talk about today, this one has a little bit of history that makes it even more interesting. Up to now, Stack Overflow has been a major loser when it comes to AI. Between October 2021 and July 2023, Stack Overflow has lost about 56 % of the traffic to its website.

10:05Stack Overflow is best known as a resource that developers use to get answers to their coding questions. Now, already you might be able to guess why the rise of AI might impact how much traffic was going to Stack Overflow. However, if you look back at seminal events from that October 2021 period, one of the big ones was GitHub Copilot. GitHub Copilot is, of course, an AI-powered tool that helped programmers by auto-completing code. It was built out on GPT-3 and began what has become a major trend and change in how people write code. In short, the more that you have AI to help write the code in the first place or support your efforts to figure out what's going wrong or debug, the less you need this sort of human-moderated forum that Stack Overflow offers.

10:48In fact, Chamath Palahapitiya actually called out this phenomenon with specific regard to Stack Overflow last year on the All In podcast. So what is Stack Overflow's AI? Well, one, it's called Overflow AI, and it basically allows people to interact with the platform via a chatbot-style interface. Remember, people have contributed 58 million answers to Stack Overflow in the past. That kind of seems like a perfect use case to bring the power of an LLM into the experience to be able to better search and sum up questions that can tap into that massive database of answers. Now, on top of that LLM-based search experience, Overflow AI also has something they call a Visual Studio Code extension, which enables AI-powered pair programming.

11:31Now, some people were skeptical of this. Mike Young responded to Stack Overflow's tweet about Overflow AI saying, This You, he added a screenshot where Stack Overflow banned the use of all generative AI when posting content to the site. At the time, they wrote, overall, because the average rate of getting correct answers from ChatGPT and other generative AI technologies is too low, the posting of answers created by ChatGPT and other generative AI technologies is substantially harmful to the site and to users who are asking questions and looking for correct answers. Mike at Grabu, however, responded in saying that makes total sense and is aligned with their announcement.

12:06They're doing semantic search similar to document Q &A, still not adding any generative AI capabilities. In my opinion, that's where the leverage for them is. There's a lot of answers out there. The key is to find the answer quick and get the right one. So to try to sum up this exchange, Stack Overflow's position on AI has been that AI-based answers to questions that are posted to Stack Overflow reduce the overall quality of answers on Stack Overflow because they're too often wrong, and nothing about that is inconsistent with their new approach to AI, which is to use LLMs to better give people access to search for answers from within their massive repository of correct answers.

12:40One of the features people really seem to be liking about it is that it actually cites the sources that it used to determine the answer that it helped create. Now, people have a sense that this might be the last gasp for Stack Overflow. Santiago tweeted, Stack Overflow is fighting back. Stack Overflow traffic has been declining for several years, but after the release of ChatGPT, things took a turn for the worse. Posting activity on the site fell the equivalent of five years in just six months. People stopped visiting the site and asking questions. Many thought this was a slow and painful death and there was no coming back.

13:07Today, their CEO announced Overflow AI. If you can't beat them, join them, right? Next up, we turn to Intel, who just had their Q2 earnings call this week. And their CEO, Pat Gelsinger, was pretty clear about how they view the AI world. As Pat put it, we're going to build AI into every platform we build. So what does that actually mean? Well, the big one is something they're calling Meteor Lake. It's Intel's first consumer chip that has a built-in neural processor for machine learning tasks. Basically, what Intel is arguing is a vision of the world in the future, where every application basically needs some sort of artificial intelligence.

13:45However, not every application is going to be able to access the cloud at all times. Because of that, people are going to need AI-centered processing right on their local devices. Gelsinger said, Across every aspect of consumer, developer, and enterprise efficiency use cases, we see that there's going to be a raft of AI enablement and those will be client-centered. Earnings season was also the context for another set of hints around AI features for the future. Spotify CEO Daniel Ek talked about a few different ways in which Spotify might introduce additional AI features. Now, already Spotify has launched their DJ feature, which is a personally curated selection of music alongside an AI-powered commentary that gives you information about the artists and tracks that you're listening to.

14:26Now, I will say personally just for a moment, this seemed kind of like a gimmick at first to me, but after having used it a bit, I actually quite like the feature. Ek seems to as well. He said, DJ is a phenomenal product. It's probably one of my personal favorites over the last few years that we have developed. And we've seen really strong consumer interactions with that. And that just talks about the ability for us to contextualize and personalize all the amazing content that we have on the Spotify platform. So another area besides just better music recommendations is podcast summaries. Ek said that this could lower the barrier to entry for consumers to check out new podcasts, which would certainly be something that I was interested in.

15:00And then, of course, you can't get away from the advertising use of AI. We talked on the brief about how Meta had seen such gains in their Reels product thanks to better AI-powered content recommendations. And Ek and Spotify are thinking about something similar as well. On that call, he said, By using generative AI in our tools here, I think you're going to be able to see that we can significantly reduce the cost that it takes for advertisers to develop new ad formats. And that obviously means that you, as an advertiser, instead of having one ad, you can imagine having thousands and tested across the Spotify networks.

15:27Lastly, as TechCrunch notes, Spotify has also filed a patent for an AI-powered text-to-speech synthesis system. Speculation is that they may be looking to go much deeper in the way that AI - powered speech comes to their system, for example, narrating audiobooks or something else entirely. Lastly, TradePub Payments is reporting that DoorDash, the food delivery startup, is apparently testing an AI chatbot in some of its markets. Bloomberg has said that the system is called Dash AI and that it allows customers to, quote, get personalized restaurant recommendations with simple text prompts. The code for Dash AI gave some examples of how users might query the system.

16:00Can you show me some highly rated and affordable dinner options nearby? What's a place that delivers burgers that also has really good salad options? Where can I get authentic Asian food I like Chinese and Thai? Now, on the one hand, I have to say of the things that we've talked about so far today, this seems the most perhaps like just trying to jump on the bandwagon, right? Specifically, given today's habits around how people use DoorDash, is this really a meaningful shift and increase in the user behavior? However, here's the counter argument. If we really move into a world where natural language becomes the default interface for talking to computers and interacting with apps, then this sort of query, can you show me some highly rated and affordable dinner options nearby, is exactly the type of thing that people are going to be asking in two or three years versus now knowing how to navigate through an app's custom rating system and filters and all the sort of menu options that will feel very old if this sort of natural language interface becomes dominant.

16:50So basically one of two things happens. Either the paradigm shifts and Dash AI looks like they got the future faster, or it doesn't. And hey, everyone else was trying this AI thing at the same time. No harm, no foul. Anyways, guys, that is going to do it for today's AI breakdown. What company's AI integrations are you most excited about? Let me know in the comments. And until next time, peace.

17:14Thank you.